Non-parametric Kruskal-Wallis Test
ANOVA is a parametric test and it assumes normality as well as homogeneity of variance. What if the assumptions fail? Here, we introduce its counterpart on the non-parametric side, the Kruskal-Wallis Test. As in the Wilcoxon two-sample test, data are replaced with their ranks without regard to the grouping.
> kruskal.test(glu~bmi.cat)
Kruskal-Wallis rank sum test
data: glu by bmi.cat
Kruskal-Wallis chi-squared = 12.7342, df = 2, p-value = 0.001717
- H0: The distribution of glucose is the same for each bmi category.
- Ha: The distribution of glucose is not the same for each bmi category.
We see that we reject the null hypothesis that the distribution of glucose is the same for each bmi category at the 0.05 α-level. (χ2 = 12.73, p-value = 0.001717).
|
Conduct an appropriate test (check the normality and equal variance assumptions) to determine if the plasma glucose concentration levels are the same for the non-diabetic individuals across different age groups. People are classified into three different age groups: group1: < 30; group2: 30-39; group3: >= 40. |
Analysis of Variance (ANOVA) and Multiple Comparisons in R (R Tutorial 4.6) MarinStatsLectures [Contents
]
Summary:
- Normality Test
- One- and Two-Sample tests: t, Wilcoxon, paired t-test
- Single factor ANOVA & Kruskal-Wallace tests
- Bartlett's Test for equal variances
Reading:
- VS Chapter 8.3
Assignment:
- Homework 3 due and Homework 4 assigned.
- Select your dataset and study question for your project, and next week bring your project proposal to class.
